Video surveillance systems for security, marketing, or the like have been installed in hotels, buildings, convenience stores, and the likes. With such surveillance systems, operations are being conducted everywhere to monitor how suspicious individuals move or how customers move on the floor. In an application for tracking a person between cameras, advance calibration of positional relationship on the floor is essential in increasing accuracy. However, if such calibration needs to be conducted manually, the load on the surveillance agent becomes very large.
There have been techniques suggested for automatically estimating link relations between cameras based on camera images. For example, Patent Document 1 discloses a technique for estimating the link relations between cameras based on the frequency distributions of the appearance times and the disappearance times of all the surveillance objects extracted by the respective cameras. Also, Patent Document 2 discloses a technique for calculating similarities between objects based on color features or the like to determine whether the objects are the same person, and generating the link relations between cameras, for example.